Good catch. Looks good. There are other j.l.management tests that
force GC to happen and I'm not sure if any of them requires similar
fix. Might be worth to check if any of them has similar assumption.
Mandy
On 5/28/13 1:11 AM, Staffan Larsen wrote:
Please review the following fix:
webrev: http://cr.openjdk.java.net/~sla/8015440/webrev.00/
bug: http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=8015440
The problem is that serial gc does not compact the heap at every full gc but only every
4th full gc (by default). When the heap isn't compacted the "usage" is returned
as a higher-than-expect value. This can be changed with -XX:MarkSweepAlwaysCompactCount=1.
Testing: I have rerun the test on a machine where it failed to make sure that
it runs correctly with the extra flag.
Thanks,
/Staffan